Peat Prize GIS data service – short description

 

Dear Colleagues,

 

This service’s objective is to support you with additional GIS data and data products we made so far. For this reason it is going to be refreshed time by time, so please visit us regularly!

 

We provide here the following datasets:

 

-          Landsat images in different composites and computed indices

-          MODIS data based raster layers considered to be relevant for our mapping purposes

-          Results of Landsat image time-series classification identifying the most relevant land cover type changes since 1972.

 

All data layers are downloadable in a single zip file by clicking HERE. The map service is available here.

 

Description of data

 

This set of data consists Landsat image composites acquired under clear sky conditions. That is the reason for they are not too many in spite of the fact that we browsed all Landsat missions. The nomenclature of these images follows the rules below.

 

Layer name example: 745_comp_L5_19980727

 

The first three characters reflect the image bands for the composite according to band assignment used on the source Landsat spacecraft. From character 5-11 we coded what Landsat mission the image belongs to. From character 13 to 20 the file name contains the date of acquisition.

 

The layers, which names include NDVI or NDII abbreviations represent the layers of these two indices.

 

For NDVI it is the standard (NIR-RED)/(NIR+RED) quotient.

For NDII it is the following quotient (NIR-SWIR)/(NIR+SWIR).

 

From MODIS products the database consists the standardized and quality checked annual averages of NDVI and NDII values. For missing years we did not find full coverage passing the quality check. These layers are named like this: ndvi_2006_wgs.tif, or ndii_2010_wgs.tif.

 

The other MODIS products represent fire occurrences on an annual sum basis. I summarized each year the appearance of pixels with nominal or high confidence of fire. Because of this product consists an 8-day aggregated value the layers for each year represent number of weeks when satellites observed fire on a given area. We made an aggregated data layer summarizing the fire occurrences in the past 16 years. Its values are ranging from 0-12. In this layer value 6 means that in the past 16 years they are 6 weekly observations indicating to be set on fire.

 

Last but not least we provide two maps of land use category changes based on the classification of Landsat images.

In the file name: lulucf_rtree_fin_wgs_1.tif or lulucf_knn_fin_wgs_1.tif we indicated the classifier used for deriving the map.

The attribute table contains all the all the observed land use category combinations cca.: 4600 unique values but I named only those that have at least as big area as 50 pixels (approximately 18 ha). The two classifications were combined. This is the reason for there are smaller areas than 18 ha also classified: Suppose we have a raster combination such as 11300300 as a LULUCF code in the rtree (random forest) classified image in a larger frequency e.g.: 68 pixels count for it. In this case I check it and it got a name such as Forest (1) to Bare land (3) in 1990 and later to Other land use (0) in the next image dated back to 1996. When I join this table with the kNN classifier result table this raster combination – 11300300 – occurs perhaps more rarely e.g.: only 32 pixels count for it. In these cases I kept the combinations, too.

 

About the nomenclature of LULUCF rasters

 

Before the classification I selected the images subject for classification. I chose images from the following years having the least cloud cover: 1972, 1990, 1996, 1998, 2001, 2007, 2010, and 2015. These are 8 observations. The selected images were classified using eCognition Developer to catch the main land cover types easy to distinguish without field reference data: Forest – 1; Bare land surface – 3; River – 5; Floodplain – 6; and other (mostly cultivated land use) – 0. This last one consists more, originally separated land cover types, but mainly croplands and shrubs.

Thus, I had a result like this table for all the pixels of the test site

 

19972

1990

1996

1998

2001

2007

2010

2015

Forest

Forest

Bare

Bare

Other

Other

Bare

Other

1

1

3

3

0

0

3

0

0

0

0

3

3

0

3

0

1

1

1

1

1

1

1

3

6

6

6

6

6

5

6

6

5

6

5

6

5

5

5

0

 

By visual interpretation I categorized the most frequent codes into classes:

 

-          Forest forever

-          Floodplain forever

-          Other land forever

-          Forest to other land after 1996

-          etc.

These names correspond the numbers in the column „Category”.

 

In the case of more than one changes e.g.: 11133000 I used more detailed description (check the table above for years!): Forest to bare land after 1996 and later bare land to other after 2001. It looks like this in the table: For-Bare 96 Bare-O 01.

 

Following this logic you can figure out the other names.

 

This classification can be refined if you want. Suggestions are welcome.

I suggest to use this layer together with fire occurrences to determine the sampling stratification, because land use and fire have to major impacts on peat layer extent and thickness.

 

“Dear Colleagues,

The TanDEM-X data we requested for our work is available in the following items:

TanDEM-X data

By downloading the data you accept the the terms and conditions of User License of TanDEM-X data for Scientific Use."

 

 

Sincerely,

Gábor